CT Radiomics Analysis For Risk Prediction Of Cervical Lymph Node Metastasis In Papillary Thyroid Carcinoma

Objective:To develop and validate radiomics model based on computed tomography(CT)for preoperative prediction of lymph nodes(LN)status in patients with papillary thyroid carcinoma(PTC).Methods:A retrospective analysis of 171 patients with PTC admitted to the thyroid surgery department of China-Japan Union Hospital of Jilin University from January 2015 to February 2019 and December 2016 to December 2019 in the thyroid surgery department of Shenzhen Bao’an District Hospital,a total of 521 patients Pathologically confirmed LNs(positive: 282 cases,negative: 239 cases)were included.All patients underwent unilateral or bilateral central group lymph node dissection.Half of the LN(positive: 119 cases,negative: 141 cases)were randomly selected as the training cohort,and the remaining 50% of LN(positive: 120 cases,negative: 141 cases)formed an independent verification cohort.All patients underwent cervical enhanced CT examination within 1 month before surgery,and N stage was obtained after undergoing postoperative paraffin pathological examination.Use radiomics related software(Artificial Intelligence Kit,ITK-SNAP)to process the LNs CT images.The patient’s clinical pathological characteristics and neck-enhanced CT arterial phase images were collected,and the region of interest(ROI)in two-dimensional was manually drawn along the boundaries of each LN by two physicians with more than three years of experience in reading neck CT,using the principle of “double-blind”.A total of 396 radiomic features were extracted from each ROI.Kruskal-Wallis test and Least absolute shrinkage and selection operator(LASSO)were used to reduce the dimensionality of feature values,and further selected features which were valuable for identifying benign and malignant LNs.Multivariate Logistic regression analysis was used to establish radiomics score(Rad-score).At the same time,we obtained the preoperative diagnosis results of the traditional CT report for LN metastasis of the enrolled patients in the central neck.Finally,we combined Rad-score and traditional CT methods,using multiple logistic regression analysis to establish a prediction model.In the training cohort and the validation cohort,the receiver operating characteristic(ROC)curve,the calibration curve and the decision curve analysis(DCA)with the Delong test are used for the evaluation of identification ability,calibration ability and clinical application of Rad-score.The sensitivity,specificity,accuracy and area under the curve(AUC)data were used to compare the preoperative prediction performance of traditional CT methods,Rad-score and the combined model.Delong test was used to compare the differences in AUC of different models in the two cohorts.Results:Radiomics showed some significant association with the status of LN for eight particular features,using a cut-off of P <0.05 for both training and validation cohorts.Rad-score showed good discriminating performance with Area Under the ROC Curve(AUC)of 0.859(95% CI,0.814 to 0.905)and 0.844(95% CI,0.796 to 0.892)in the training and validation cohorts,respectively,yielding an accuracy of 81.2%(sensitivity,73.9%;specificity,87.2%)in the training cohort and 78.5%(sensitivity,66.7%;specificity,88.7%)in the validation cohort.AUC of CT was 0.689(95% CI,0.634 to 0.744)in training cohort,and 0.631(95% CI,0.575 to 0.687)in validation cohort.Significant differences between the AUCs of conventional CT-reported LN status& Rad-score,and conventional CT & the combined model in both training and validation cohorts were noticed(P < 0.05).However,there was no significant difference between Rad-score and the combined model of AUC(P < 0.05).In the training and validation cohorts,the calibration curves drawn for Rad-score showed good agreement between the predicted results and the final observation results,and the Hosmer-Lemeshow test P-values of the training cohort and validation cohort were0.471 and 0.130,respectively,indicating the applicability of Rad-score.DCA showed that when a certain threshold probability is reached,Rad-score adds more clinical net benefits than conventional CT.Conclusion:1.CT radiomics showed a favorable performance for the prediction of cervical lymph node metastasis in PTC.2.The CT radiomics model outperformed conventional CT in predicting cervical lymph node metastasis in PTC.3.The combined model combined with CT radiomics and conventional CT report showed highly sensitive to predict cervical lymph node metastasis in PTC.